falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
24554 Commits
32 Branches
283MB
Tree: 44c70a9b3d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '44c70a9b3d'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2639 lines
79KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

  3.  * Copyright (c) 2001, 2002 Fabrice Bellard

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

  7.  * FFmpeg is free software; you can redistribute it and/or

  8.  * modify it under the terms of the GNU Lesser General Public

  9.  * License as published by the Free Software Foundation; either

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * FFmpeg is distributed in the hope that it will be useful,

  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  15.  * Lesser General Public License for more details.

  16.  *

  17.  * You should have received a copy of the GNU Lesser General Public

  18.  * License along with FFmpeg; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * MPEG Audio decoder.

  25.  */

  26. 

  27. #include "avcodec.h"

  28. #include "get_bits.h"

  29. #include "dsputil.h"

  30. 

  31. /*

  32.  * TODO:

  33.  *  - in low precision mode, use more 16 bit multiplies in synth filter

  34.  *  - test lsf / mpeg25 extensively.

  35.  */

  36. 

  37. #include "mpegaudio.h"

  38. #include "mpegaudiodecheader.h"

  39. 

  40. #include "mathops.h"

  41. 

  42. #if CONFIG_FLOAT

  43. #   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))

  44. #   define compute_antialias compute_antialias_float

  45. #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))

  46. #   define FIXR(x)        ((float)(x))

  47. #   define FIXHR(x)       ((float)(x))

  48. #   define MULH3(x, y, s) ((s)*(y)*(x))

  49. #   define MULLx(x, y, s) ((y)*(x))

  50. #   define RENAME(a) a ## _float

  51. #else

  52. #   define SHR(a,b)       ((a)>>(b))

  53. #   define compute_antialias compute_antialias_integer

  54. /* WARNING: only correct for posititive numbers */

  55. #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))

  56. #   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))

  57. #   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))

  58. #   define MULH3(x, y, s) MULH((s)*(x), y)

  59. #   define MULLx(x, y, s) MULL(x,y,s)

  60. #   define RENAME(a)      a

  61. #endif

  62. 

  63. /****************/

  64. 

  65. #define HEADER_SIZE 4

  66. 

  67. #include "mpegaudiodata.h"

  68. #include "mpegaudiodectab.h"

  69. 

  70. static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);

  71. static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);

  72. static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,

  73.                                int *dither_state, OUT_INT *samples, int incr);

  74. 

  75. /* vlc structure for decoding layer 3 huffman tables */

  76. static VLC huff_vlc[16];

  77. static VLC_TYPE huff_vlc_tables[

  78.   0+128+128+128+130+128+154+166+

  79.   142+204+190+170+542+460+662+414

  80.   ][2];

  81. static const int huff_vlc_tables_sizes[16] = {

  82.   0, 128, 128, 128, 130, 128, 154, 166,

  83.   142, 204, 190, 170, 542, 460, 662, 414

  84. };

  85. static VLC huff_quad_vlc[2];

  86. static VLC_TYPE huff_quad_vlc_tables[128+16][2];

  87. static const int huff_quad_vlc_tables_sizes[2] = {

  88.   128, 16

  89. };

  90. /* computed from band_size_long */

  91. static uint16_t band_index_long[9][23];

  92. #include "mpegaudio_tablegen.h"

  93. /* intensity stereo coef table */

  94. static INTFLOAT is_table[2][16];

  95. static INTFLOAT is_table_lsf[2][2][16];

  96. static int32_t csa_table[8][4];

  97. static float csa_table_float[8][4];

  98. static INTFLOAT mdct_win[8][36];

  99. 

  100. /* lower 2 bits: modulo 3, higher bits: shift */

  101. static uint16_t scale_factor_modshift[64];

  102. /* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */

  103. static int32_t scale_factor_mult[15][3];

  104. /* mult table for layer 2 group quantization */

  105. 

  106. #define SCALE_GEN(v) \

  107. { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }

  108. 

  109. static const int32_t scale_factor_mult2[3][3] = {

  110.     SCALE_GEN(4.0 / 3.0), /* 3 steps */

  111.     SCALE_GEN(4.0 / 5.0), /* 5 steps */

  112.     SCALE_GEN(4.0 / 9.0), /* 9 steps */

  113. };

  114. 

  115. DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512];

  116. 

  117. /**

  118.  * Convert region offsets to region sizes and truncate

  119.  * size to big_values.

  120.  */

  121. static void ff_region_offset2size(GranuleDef *g){

  122.     int i, k, j=0;

  123.     g->region_size[2] = (576 / 2);

  124.     for(i=0;i<3;i++) {

  125.         k = FFMIN(g->region_size[i], g->big_values);

  126.         g->region_size[i] = k - j;

  127.         j = k;

  128.     }

  129. }

  130. 

  131. static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){

  132.     if (g->block_type == 2)

  133.         g->region_size[0] = (36 / 2);

  134.     else {

  135.         if (s->sample_rate_index <= 2)

  136.             g->region_size[0] = (36 / 2);

  137.         else if (s->sample_rate_index != 8)

  138.             g->region_size[0] = (54 / 2);

  139.         else

  140.             g->region_size[0] = (108 / 2);

  141.     }

  142.     g->region_size[1] = (576 / 2);

  143. }

  144. 

  145. static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){

  146.     int l;

  147.     g->region_size[0] =

  148.         band_index_long[s->sample_rate_index][ra1 + 1] >> 1;

  149.     /* should not overflow */

  150.     l = FFMIN(ra1 + ra2 + 2, 22);

  151.     g->region_size[1] =

  152.         band_index_long[s->sample_rate_index][l] >> 1;

  153. }

  154. 

  155. static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){

  156.     if (g->block_type == 2) {

  157.         if (g->switch_point) {

  158.             /* if switched mode, we handle the 36 first samples as

  159.                 long blocks.  For 8000Hz, we handle the 48 first

  160.                 exponents as long blocks (XXX: check this!) */

  161.             if (s->sample_rate_index <= 2)

  162.                 g->long_end = 8;

  163.             else if (s->sample_rate_index != 8)

  164.                 g->long_end = 6;

  165.             else

  166.                 g->long_end = 4; /* 8000 Hz */

  167. 

  168.             g->short_start = 2 + (s->sample_rate_index != 8);

  169.         } else {

  170.             g->long_end = 0;

  171.             g->short_start = 0;

  172.         }

  173.     } else {

  174.         g->short_start = 13;

  175.         g->long_end = 22;

  176.     }

  177. }

  178. 

  179. /* layer 1 unscaling */

  180. /* n = number of bits of the mantissa minus 1 */

  181. static inline int l1_unscale(int n, int mant, int scale_factor)

  182. {

  183.     int shift, mod;

  184.     int64_t val;

  185. 

  186.     shift = scale_factor_modshift[scale_factor];

  187.     mod = shift & 3;

  188.     shift >>= 2;

  189.     val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);

  190.     shift += n;

  191.     /* NOTE: at this point, 1 <= shift >= 21 + 15 */

  192.     return (int)((val + (1LL << (shift - 1))) >> shift);

  193. }

  194. 

  195. static inline int l2_unscale_group(int steps, int mant, int scale_factor)

  196. {

  197.     int shift, mod, val;

  198. 

  199.     shift = scale_factor_modshift[scale_factor];

  200.     mod = shift & 3;

  201.     shift >>= 2;

  202. 

  203.     val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];

  204.     /* NOTE: at this point, 0 <= shift <= 21 */

  205.     if (shift > 0)

  206.         val = (val + (1 << (shift - 1))) >> shift;

  207.     return val;

  208. }

  209. 

  210. /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */

  211. static inline int l3_unscale(int value, int exponent)

  212. {

  213.     unsigned int m;

  214.     int e;

  215. 

  216.     e = table_4_3_exp  [4*value + (exponent&3)];

  217.     m = table_4_3_value[4*value + (exponent&3)];

  218.     e -= (exponent >> 2);

  219.     assert(e>=1);

  220.     if (e > 31)

  221.         return 0;

  222.     m = (m + (1 << (e-1))) >> e;

  223. 

  224.     return m;

  225. }

  226. 

  227. /* all integer n^(4/3) computation code */

  228. #define DEV_ORDER 13

  229. 

  230. #define POW_FRAC_BITS 24

  231. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

  232. #define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))

  233. #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)

  234. 

  235. static int dev_4_3_coefs[DEV_ORDER];

  236. 

  237. #if 0 /* unused */

  238. static int pow_mult3[3] = {

  239.     POW_FIX(1.0),

  240.     POW_FIX(1.25992104989487316476),

  241.     POW_FIX(1.58740105196819947474),

  242. };

  243. #endif

  244. 

  245. static av_cold void int_pow_init(void)

  246. {

  247.     int i, a;

  248. 

  249.     a = POW_FIX(1.0);

  250.     for(i=0;i<DEV_ORDER;i++) {

  251.         a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);

  252.         dev_4_3_coefs[i] = a;

  253.     }

  254. }

  255. 

  256. #if 0 /* unused, remove? */

  257. /* return the mantissa and the binary exponent */

  258. static int int_pow(int i, int *exp_ptr)

  259. {

  260.     int e, er, eq, j;

  261.     int a, a1;

  262. 

  263.     /* renormalize */

  264.     a = i;

  265.     e = POW_FRAC_BITS;

  266.     while (a < (1 << (POW_FRAC_BITS - 1))) {

  267.         a = a << 1;

  268.         e--;

  269.     }

  270.     a -= (1 << POW_FRAC_BITS);

  271.     a1 = 0;

  272.     for(j = DEV_ORDER - 1; j >= 0; j--)

  273.         a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);

  274.     a = (1 << POW_FRAC_BITS) + a1;

  275.     /* exponent compute (exact) */

  276.     e = e * 4;

  277.     er = e % 3;

  278.     eq = e / 3;

  279.     a = POW_MULL(a, pow_mult3[er]);

  280.     while (a >= 2 * POW_FRAC_ONE) {

  281.         a = a >> 1;

  282.         eq++;

  283.     }

  284.     /* convert to float */

  285.     while (a < POW_FRAC_ONE) {

  286.         a = a << 1;

  287.         eq--;

  288.     }

  289.     /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */

  290. #if POW_FRAC_BITS > FRAC_BITS

  291.     a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);

  292.     /* correct overflow */

  293.     if (a >= 2 * (1 << FRAC_BITS)) {

  294.         a = a >> 1;

  295.         eq++;

  296.     }

  297. #endif

  298.     *exp_ptr = eq;

  299.     return a;

  300. }

  301. #endif

  302. 

  303. static av_cold int decode_init(AVCodecContext * avctx)

  304. {

  305.     MPADecodeContext *s = avctx->priv_data;

  306.     static int init=0;

  307.     int i, j, k;

  308. 

  309.     s->avctx = avctx;

  310.     s->apply_window_mp3 = apply_window_mp3_c;

  311. 

  312.     avctx->sample_fmt= OUT_FMT;

  313.     s->error_recognition= avctx->error_recognition;

  314. 

  315.     if (!init && !avctx->parse_only) {

  316.         int offset;

  317. 

  318.         /* scale factors table for layer 1/2 */

  319.         for(i=0;i<64;i++) {

  320.             int shift, mod;

  321.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  322.             shift = (i / 3);

  323.             mod = i % 3;

  324.             scale_factor_modshift[i] = mod | (shift << 2);

  325.         }

  326. 

  327.         /* scale factor multiply for layer 1 */

  328.         for(i=0;i<15;i++) {

  329.             int n, norm;

  330.             n = i + 2;

  331.             norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);

  332.             scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0          * 2.0), FRAC_BITS);

  333.             scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);

  334.             scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);

  335.             dprintf(avctx, "%d: norm=%x s=%x %x %x\n",

  336.                     i, norm,

  337.                     scale_factor_mult[i][0],

  338.                     scale_factor_mult[i][1],

  339.                     scale_factor_mult[i][2]);

  340.         }

  341. 

  342.         RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));

  343. 

  344.         /* huffman decode tables */

  345.         offset = 0;

  346.         for(i=1;i<16;i++) {

  347.             const HuffTable *h = &mpa_huff_tables[i];

  348.             int xsize, x, y;

  349.             uint8_t  tmp_bits [512];

  350.             uint16_t tmp_codes[512];

  351. 

  352.             memset(tmp_bits , 0, sizeof(tmp_bits ));

  353.             memset(tmp_codes, 0, sizeof(tmp_codes));

  354. 

  355.             xsize = h->xsize;

  356. 

  357.             j = 0;

  358.             for(x=0;x<xsize;x++) {

  359.                 for(y=0;y<xsize;y++){

  360.                     tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];

  361.                     tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];

  362.                 }

  363.             }

  364. 

  365.             /* XXX: fail test */

  366.             huff_vlc[i].table = huff_vlc_tables+offset;

  367.             huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];

  368.             init_vlc(&huff_vlc[i], 7, 512,

  369.                      tmp_bits, 1, 1, tmp_codes, 2, 2,

  370.                      INIT_VLC_USE_NEW_STATIC);

  371.             offset += huff_vlc_tables_sizes[i];

  372.         }

  373.         assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));

  374. 

  375.         offset = 0;

  376.         for(i=0;i<2;i++) {

  377.             huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;

  378.             huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];

  379.             init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,

  380.                      mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,

  381.                      INIT_VLC_USE_NEW_STATIC);

  382.             offset += huff_quad_vlc_tables_sizes[i];

  383.         }

  384.         assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));

  385. 

  386.         for(i=0;i<9;i++) {

  387.             k = 0;

  388.             for(j=0;j<22;j++) {

  389.                 band_index_long[i][j] = k;

  390.                 k += band_size_long[i][j];

  391.             }

  392.             band_index_long[i][22] = k;

  393.         }

  394. 

  395.         /* compute n ^ (4/3) and store it in mantissa/exp format */

  396. 

  397.         int_pow_init();

  398.         mpegaudio_tableinit();

  399. 

  400.         for(i=0;i<7;i++) {

  401.             float f;

  402.             INTFLOAT v;

  403.             if (i != 6) {

  404.                 f = tan((double)i * M_PI / 12.0);

  405.                 v = FIXR(f / (1.0 + f));

  406.             } else {

  407.                 v = FIXR(1.0);

  408.             }

  409.             is_table[0][i] = v;

  410.             is_table[1][6 - i] = v;

  411.         }

  412.         /* invalid values */

  413.         for(i=7;i<16;i++)

  414.             is_table[0][i] = is_table[1][i] = 0.0;

  415. 

  416.         for(i=0;i<16;i++) {

  417.             double f;

  418.             int e, k;

  419. 

  420.             for(j=0;j<2;j++) {

  421.                 e = -(j + 1) * ((i + 1) >> 1);

  422.                 f = pow(2.0, e / 4.0);

  423.                 k = i & 1;

  424.                 is_table_lsf[j][k ^ 1][i] = FIXR(f);

  425.                 is_table_lsf[j][k][i] = FIXR(1.0);

  426.                 dprintf(avctx, "is_table_lsf %d %d: %x %x\n",

  427.                         i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);

  428.             }

  429.         }

  430. 

  431.         for(i=0;i<8;i++) {

  432.             float ci, cs, ca;

  433.             ci = ci_table[i];

  434.             cs = 1.0 / sqrt(1.0 + ci * ci);

  435.             ca = cs * ci;

  436.             csa_table[i][0] = FIXHR(cs/4);

  437.             csa_table[i][1] = FIXHR(ca/4);

  438.             csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);

  439.             csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);

  440.             csa_table_float[i][0] = cs;

  441.             csa_table_float[i][1] = ca;

  442.             csa_table_float[i][2] = ca + cs;

  443.             csa_table_float[i][3] = ca - cs;

  444.         }

  445. 

  446.         /* compute mdct windows */

  447.         for(i=0;i<36;i++) {

  448.             for(j=0; j<4; j++){

  449.                 double d;

  450. 

  451.                 if(j==2 && i%3 != 1)

  452.                     continue;

  453. 

  454.                 d= sin(M_PI * (i + 0.5) / 36.0);

  455.                 if(j==1){

  456.                     if     (i>=30) d= 0;

  457.                     else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);

  458.                     else if(i>=18) d= 1;

  459.                 }else if(j==3){

  460.                     if     (i<  6) d= 0;

  461.                     else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);

  462.                     else if(i< 18) d= 1;

  463.                 }

  464.                 //merge last stage of imdct into the window coefficients

  465.                 d*= 0.5 / cos(M_PI*(2*i + 19)/72);

  466. 

  467.                 if(j==2)

  468.                     mdct_win[j][i/3] = FIXHR((d / (1<<5)));

  469.                 else

  470.                     mdct_win[j][i  ] = FIXHR((d / (1<<5)));

  471.             }

  472.         }

  473. 

  474.         /* NOTE: we do frequency inversion adter the MDCT by changing

  475.            the sign of the right window coefs */

  476.         for(j=0;j<4;j++) {

  477.             for(i=0;i<36;i+=2) {

  478.                 mdct_win[j + 4][i] = mdct_win[j][i];

  479.                 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];

  480.             }

  481.         }

  482. 

  483.         init = 1;

  484.     }

  485. 

  486.     if (avctx->codec_id == CODEC_ID_MP3ADU)

  487.         s->adu_mode = 1;

  488.     return 0;

  489. }

  490. 

  491. /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */

  492. 

  493. /* cos(i*pi/64) */

  494. 

  495. #define COS0_0  FIXHR(0.50060299823519630134/2)

  496. #define COS0_1  FIXHR(0.50547095989754365998/2)

  497. #define COS0_2  FIXHR(0.51544730992262454697/2)

  498. #define COS0_3  FIXHR(0.53104259108978417447/2)

  499. #define COS0_4  FIXHR(0.55310389603444452782/2)

  500. #define COS0_5  FIXHR(0.58293496820613387367/2)

  501. #define COS0_6  FIXHR(0.62250412303566481615/2)

  502. #define COS0_7  FIXHR(0.67480834145500574602/2)

  503. #define COS0_8  FIXHR(0.74453627100229844977/2)

  504. #define COS0_9  FIXHR(0.83934964541552703873/2)

  505. #define COS0_10 FIXHR(0.97256823786196069369/2)

  506. #define COS0_11 FIXHR(1.16943993343288495515/4)

  507. #define COS0_12 FIXHR(1.48416461631416627724/4)

  508. #define COS0_13 FIXHR(2.05778100995341155085/8)

  509. #define COS0_14 FIXHR(3.40760841846871878570/8)

  510. #define COS0_15 FIXHR(10.19000812354805681150/32)

  511. 

  512. #define COS1_0 FIXHR(0.50241928618815570551/2)

  513. #define COS1_1 FIXHR(0.52249861493968888062/2)

  514. #define COS1_2 FIXHR(0.56694403481635770368/2)

  515. #define COS1_3 FIXHR(0.64682178335999012954/2)

  516. #define COS1_4 FIXHR(0.78815462345125022473/2)

  517. #define COS1_5 FIXHR(1.06067768599034747134/4)

  518. #define COS1_6 FIXHR(1.72244709823833392782/4)

  519. #define COS1_7 FIXHR(5.10114861868916385802/16)

  520. 

  521. #define COS2_0 FIXHR(0.50979557910415916894/2)

  522. #define COS2_1 FIXHR(0.60134488693504528054/2)

  523. #define COS2_2 FIXHR(0.89997622313641570463/2)

  524. #define COS2_3 FIXHR(2.56291544774150617881/8)

  525. 

  526. #define COS3_0 FIXHR(0.54119610014619698439/2)

  527. #define COS3_1 FIXHR(1.30656296487637652785/4)

  528. 

  529. #define COS4_0 FIXHR(0.70710678118654752439/2)

  530. 

  531. /* butterfly operator */

  532. #define BF(a, b, c, s)\

  533. {\

  534.     tmp0 = val##a + val##b;\

  535.     tmp1 = val##a - val##b;\

  536.     val##a = tmp0;\

  537.     val##b = MULH3(tmp1, c, 1<<(s));\

  538. }

  539. 

  540. #define BF0(a, b, c, s)\

  541. {\

  542.     tmp0 = tab[a] + tab[b];\

  543.     tmp1 = tab[a] - tab[b];\

  544.     val##a = tmp0;\

  545.     val##b = MULH3(tmp1, c, 1<<(s));\

  546. }

  547. 

  548. #define BF1(a, b, c, d)\

  549. {\

  550.     BF(a, b, COS4_0, 1);\

  551.     BF(c, d,-COS4_0, 1);\

  552.     val##c += val##d;\

  553. }

  554. 

  555. #define BF2(a, b, c, d)\

  556. {\

  557.     BF(a, b, COS4_0, 1);\

  558.     BF(c, d,-COS4_0, 1);\

  559.     val##c += val##d;\

  560.     val##a += val##c;\

  561.     val##c += val##b;\

  562.     val##b += val##d;\

  563. }

  564. 

  565. #define ADD(a, b) val##a += val##b

  566. 

  567. /* DCT32 without 1/sqrt(2) coef zero scaling. */

  568. static void dct32(INTFLOAT *out, const INTFLOAT *tab)

  569. {

  570.     INTFLOAT tmp0, tmp1;

  571. 

  572.     INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 ,

  573.              val8 , val9 , val10, val11, val12, val13, val14, val15,

  574.              val16, val17, val18, val19, val20, val21, val22, val23,

  575.              val24, val25, val26, val27, val28, val29, val30, val31;

  576. 

  577.     /* pass 1 */

  578.     BF0( 0, 31, COS0_0 , 1);

  579.     BF0(15, 16, COS0_15, 5);

  580.     /* pass 2 */

  581.     BF( 0, 15, COS1_0 , 1);

  582.     BF(16, 31,-COS1_0 , 1);

  583.     /* pass 1 */

  584.     BF0( 7, 24, COS0_7 , 1);

  585.     BF0( 8, 23, COS0_8 , 1);

  586.     /* pass 2 */

  587.     BF( 7,  8, COS1_7 , 4);

  588.     BF(23, 24,-COS1_7 , 4);

  589.     /* pass 3 */

  590.     BF( 0,  7, COS2_0 , 1);

  591.     BF( 8, 15,-COS2_0 , 1);

  592.     BF(16, 23, COS2_0 , 1);

  593.     BF(24, 31,-COS2_0 , 1);

  594.     /* pass 1 */

  595.     BF0( 3, 28, COS0_3 , 1);

  596.     BF0(12, 19, COS0_12, 2);

  597.     /* pass 2 */

  598.     BF( 3, 12, COS1_3 , 1);

  599.     BF(19, 28,-COS1_3 , 1);

  600.     /* pass 1 */

  601.     BF0( 4, 27, COS0_4 , 1);

  602.     BF0(11, 20, COS0_11, 2);

  603.     /* pass 2 */

  604.     BF( 4, 11, COS1_4 , 1);

  605.     BF(20, 27,-COS1_4 , 1);

  606.     /* pass 3 */

  607.     BF( 3,  4, COS2_3 , 3);

  608.     BF(11, 12,-COS2_3 , 3);

  609.     BF(19, 20, COS2_3 , 3);

  610.     BF(27, 28,-COS2_3 , 3);

  611.     /* pass 4 */

  612.     BF( 0,  3, COS3_0 , 1);

  613.     BF( 4,  7,-COS3_0 , 1);

  614.     BF( 8, 11, COS3_0 , 1);

  615.     BF(12, 15,-COS3_0 , 1);

  616.     BF(16, 19, COS3_0 , 1);

  617.     BF(20, 23,-COS3_0 , 1);

  618.     BF(24, 27, COS3_0 , 1);

  619.     BF(28, 31,-COS3_0 , 1);

  620. 

  621. 

  622. 

  623.     /* pass 1 */

  624.     BF0( 1, 30, COS0_1 , 1);

  625.     BF0(14, 17, COS0_14, 3);

  626.     /* pass 2 */

  627.     BF( 1, 14, COS1_1 , 1);

  628.     BF(17, 30,-COS1_1 , 1);

  629.     /* pass 1 */

  630.     BF0( 6, 25, COS0_6 , 1);

  631.     BF0( 9, 22, COS0_9 , 1);

  632.     /* pass 2 */

  633.     BF( 6,  9, COS1_6 , 2);

  634.     BF(22, 25,-COS1_6 , 2);

  635.     /* pass 3 */

  636.     BF( 1,  6, COS2_1 , 1);

  637.     BF( 9, 14,-COS2_1 , 1);

  638.     BF(17, 22, COS2_1 , 1);

  639.     BF(25, 30,-COS2_1 , 1);

  640. 

  641.     /* pass 1 */

  642.     BF0( 2, 29, COS0_2 , 1);

  643.     BF0(13, 18, COS0_13, 3);

  644.     /* pass 2 */

  645.     BF( 2, 13, COS1_2 , 1);

  646.     BF(18, 29,-COS1_2 , 1);

  647.     /* pass 1 */

  648.     BF0( 5, 26, COS0_5 , 1);

  649.     BF0(10, 21, COS0_10, 1);

  650.     /* pass 2 */

  651.     BF( 5, 10, COS1_5 , 2);

  652.     BF(21, 26,-COS1_5 , 2);

  653.     /* pass 3 */

  654.     BF( 2,  5, COS2_2 , 1);

  655.     BF(10, 13,-COS2_2 , 1);

  656.     BF(18, 21, COS2_2 , 1);

  657.     BF(26, 29,-COS2_2 , 1);

  658.     /* pass 4 */

  659.     BF( 1,  2, COS3_1 , 2);

  660.     BF( 5,  6,-COS3_1 , 2);

  661.     BF( 9, 10, COS3_1 , 2);

  662.     BF(13, 14,-COS3_1 , 2);

  663.     BF(17, 18, COS3_1 , 2);

  664.     BF(21, 22,-COS3_1 , 2);

  665.     BF(25, 26, COS3_1 , 2);

  666.     BF(29, 30,-COS3_1 , 2);

  667. 

  668.     /* pass 5 */

  669.     BF1( 0,  1,  2,  3);

  670.     BF2( 4,  5,  6,  7);

  671.     BF1( 8,  9, 10, 11);

  672.     BF2(12, 13, 14, 15);

  673.     BF1(16, 17, 18, 19);

  674.     BF2(20, 21, 22, 23);

  675.     BF1(24, 25, 26, 27);

  676.     BF2(28, 29, 30, 31);

  677. 

  678.     /* pass 6 */

  679. 

  680.     ADD( 8, 12);

  681.     ADD(12, 10);

  682.     ADD(10, 14);

  683.     ADD(14,  9);

  684.     ADD( 9, 13);

  685.     ADD(13, 11);

  686.     ADD(11, 15);

  687. 

  688.     out[ 0] = val0;

  689.     out[16] = val1;

  690.     out[ 8] = val2;

  691.     out[24] = val3;

  692.     out[ 4] = val4;

  693.     out[20] = val5;

  694.     out[12] = val6;

  695.     out[28] = val7;

  696.     out[ 2] = val8;

  697.     out[18] = val9;

  698.     out[10] = val10;

  699.     out[26] = val11;

  700.     out[ 6] = val12;

  701.     out[22] = val13;

  702.     out[14] = val14;

  703.     out[30] = val15;

  704. 

  705.     ADD(24, 28);

  706.     ADD(28, 26);

  707.     ADD(26, 30);

  708.     ADD(30, 25);

  709.     ADD(25, 29);

  710.     ADD(29, 27);

  711.     ADD(27, 31);

  712. 

  713.     out[ 1] = val16 + val24;

  714.     out[17] = val17 + val25;

  715.     out[ 9] = val18 + val26;

  716.     out[25] = val19 + val27;

  717.     out[ 5] = val20 + val28;

  718.     out[21] = val21 + val29;

  719.     out[13] = val22 + val30;

  720.     out[29] = val23 + val31;

  721.     out[ 3] = val24 + val20;

  722.     out[19] = val25 + val21;

  723.     out[11] = val26 + val22;

  724.     out[27] = val27 + val23;

  725.     out[ 7] = val28 + val18;

  726.     out[23] = val29 + val19;

  727.     out[15] = val30 + val17;

  728.     out[31] = val31;

  729. }

  730. 

  731. #if CONFIG_FLOAT

  732. static inline float round_sample(float *sum)

  733. {

  734.     float sum1=*sum;

  735.     *sum = 0;

  736.     return sum1;

  737. }

  738. 

  739. /* signed 16x16 -> 32 multiply add accumulate */

  740. #define MACS(rt, ra, rb) rt+=(ra)*(rb)

  741. 

  742. /* signed 16x16 -> 32 multiply */

  743. #define MULS(ra, rb) ((ra)*(rb))

  744. 

  745. #define MLSS(rt, ra, rb) rt-=(ra)*(rb)

  746. 

  747. #elif FRAC_BITS <= 15

  748. 

  749. static inline int round_sample(int *sum)

  750. {

  751.     int sum1;

  752.     sum1 = (*sum) >> OUT_SHIFT;

  753.     *sum &= (1<<OUT_SHIFT)-1;

  754.     return av_clip(sum1, OUT_MIN, OUT_MAX);

  755. }

  756. 

  757. /* signed 16x16 -> 32 multiply add accumulate */

  758. #define MACS(rt, ra, rb) MAC16(rt, ra, rb)

  759. 

  760. /* signed 16x16 -> 32 multiply */

  761. #define MULS(ra, rb) MUL16(ra, rb)

  762. 

  763. #define MLSS(rt, ra, rb) MLS16(rt, ra, rb)

  764. 

  765. #else

  766. 

  767. static inline int round_sample(int64_t *sum)

  768. {

  769.     int sum1;

  770.     sum1 = (int)((*sum) >> OUT_SHIFT);

  771.     *sum &= (1<<OUT_SHIFT)-1;

  772.     return av_clip(sum1, OUT_MIN, OUT_MAX);

  773. }

  774. 

  775. #   define MULS(ra, rb) MUL64(ra, rb)

  776. #   define MACS(rt, ra, rb) MAC64(rt, ra, rb)

  777. #   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)

  778. #endif

  779. 

  780. #define SUM8(op, sum, w, p)               \

  781. {                                         \

  782.     op(sum, (w)[0 * 64], (p)[0 * 64]);    \

  783.     op(sum, (w)[1 * 64], (p)[1 * 64]);    \

  784.     op(sum, (w)[2 * 64], (p)[2 * 64]);    \

  785.     op(sum, (w)[3 * 64], (p)[3 * 64]);    \

  786.     op(sum, (w)[4 * 64], (p)[4 * 64]);    \

  787.     op(sum, (w)[5 * 64], (p)[5 * 64]);    \

  788.     op(sum, (w)[6 * 64], (p)[6 * 64]);    \

  789.     op(sum, (w)[7 * 64], (p)[7 * 64]);    \

  790. }

  791. 

  792. #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \

  793. {                                               \

  794.     INTFLOAT tmp;\

  795.     tmp = p[0 * 64];\

  796.     op1(sum1, (w1)[0 * 64], tmp);\

  797.     op2(sum2, (w2)[0 * 64], tmp);\

  798.     tmp = p[1 * 64];\

  799.     op1(sum1, (w1)[1 * 64], tmp);\

  800.     op2(sum2, (w2)[1 * 64], tmp);\

  801.     tmp = p[2 * 64];\

  802.     op1(sum1, (w1)[2 * 64], tmp);\

  803.     op2(sum2, (w2)[2 * 64], tmp);\

  804.     tmp = p[3 * 64];\

  805.     op1(sum1, (w1)[3 * 64], tmp);\

  806.     op2(sum2, (w2)[3 * 64], tmp);\

  807.     tmp = p[4 * 64];\

  808.     op1(sum1, (w1)[4 * 64], tmp);\

  809.     op2(sum2, (w2)[4 * 64], tmp);\

  810.     tmp = p[5 * 64];\

  811.     op1(sum1, (w1)[5 * 64], tmp);\

  812.     op2(sum2, (w2)[5 * 64], tmp);\

  813.     tmp = p[6 * 64];\

  814.     op1(sum1, (w1)[6 * 64], tmp);\

  815.     op2(sum2, (w2)[6 * 64], tmp);\

  816.     tmp = p[7 * 64];\

  817.     op1(sum1, (w1)[7 * 64], tmp);\

  818.     op2(sum2, (w2)[7 * 64], tmp);\

  819. }

  820. 

  821. void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)

  822. {

  823.     int i;

  824. 

  825.     /* max = 18760, max sum over all 16 coefs : 44736 */

  826.     for(i=0;i<257;i++) {

  827.         INTFLOAT v;

  828.         v = ff_mpa_enwindow[i];

  829. #if CONFIG_FLOAT

  830.         v *= 1.0 / (1LL<<(16 + FRAC_BITS));

  831. #elif WFRAC_BITS < 16

  832.         v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);

  833. #endif

  834.         window[i] = v;

  835.         if ((i & 63) != 0)

  836.             v = -v;

  837.         if (i != 0)

  838.             window[512 - i] = v;

  839.     }

  840. }

  841. 

  842. static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,

  843.                                int *dither_state, OUT_INT *samples, int incr)

  844. {

  845.     register const MPA_INT *w, *w2, *p;

  846.     int j;

  847.     OUT_INT *samples2;

  848. #if CONFIG_FLOAT

  849.     float sum, sum2;

  850. #elif FRAC_BITS <= 15

  851.     int sum, sum2;

  852. #else

  853.     int64_t sum, sum2;

  854. #endif

  855. 

  856.     /* copy to avoid wrap */

  857.     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));

  858. 

  859.     samples2 = samples + 31 * incr;

  860.     w = window;

  861.     w2 = window + 31;

  862. 

  863.     sum = *dither_state;

  864.     p = synth_buf + 16;

  865.     SUM8(MACS, sum, w, p);

  866.     p = synth_buf + 48;

  867.     SUM8(MLSS, sum, w + 32, p);

  868.     *samples = round_sample(&sum);

  869.     samples += incr;

  870.     w++;

  871. 

  872.     /* we calculate two samples at the same time to avoid one memory

  873.        access per two sample */

  874.     for(j=1;j<16;j++) {

  875.         sum2 = 0;

  876.         p = synth_buf + 16 + j;

  877.         SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);

  878.         p = synth_buf + 48 - j;

  879.         SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);

  880. 

  881.         *samples = round_sample(&sum);

  882.         samples += incr;

  883.         sum += sum2;

  884.         *samples2 = round_sample(&sum);

  885.         samples2 -= incr;

  886.         w++;

  887.         w2--;

  888.     }

  889. 

  890.     p = synth_buf + 32;

  891.     SUM8(MLSS, sum, w + 32, p);

  892.     *samples = round_sample(&sum);

  893.     *dither_state= sum;

  894. }

  895. 

  896. 

  897. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  898.    32 samples. */

  899. /* XXX: optimize by avoiding ring buffer usage */

  900. #if CONFIG_FLOAT

  901. void ff_mpa_synth_filter_float(MPADecodeContext *s, float *synth_buf_ptr,

  902.                                int *synth_buf_offset,

  903.                                float *window, int *dither_state,

  904.                                float *samples, int incr,

  905.                                float sb_samples[SBLIMIT])

  906. {

  907.     float *synth_buf;

  908.     int offset;

  909. 

  910.     offset = *synth_buf_offset;

  911.     synth_buf = synth_buf_ptr + offset;

  912. 

  913.     dct32(synth_buf, sb_samples);

  914.     s->apply_window_mp3(synth_buf, window, dither_state, samples, incr);

  915. 

  916.     offset = (offset - 32) & 511;

  917.     *synth_buf_offset = offset;

  918. }

  919. #else

  920. void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,

  921.                          MPA_INT *window, int *dither_state,

  922.                          OUT_INT *samples, int incr,

  923.                          INTFLOAT sb_samples[SBLIMIT])

  924. {

  925.     register MPA_INT *synth_buf;

  926.     int offset;

  927. #if FRAC_BITS <= 15

  928.     int32_t tmp[32];

  929.     int j;

  930. #endif

  931. 

  932.     offset = *synth_buf_offset;

  933.     synth_buf = synth_buf_ptr + offset;

  934. 

  935. #if FRAC_BITS <= 15 && !CONFIG_FLOAT

  936.     dct32(tmp, sb_samples);

  937.     for(j=0;j<32;j++) {

  938.         /* NOTE: can cause a loss in precision if very high amplitude

  939.            sound */

  940.         synth_buf[j] = av_clip_int16(tmp[j]);

  941.     }

  942. #else

  943.     dct32(synth_buf, sb_samples);

  944. #endif

  945. 

  946.     apply_window_mp3_c(synth_buf, window, dither_state, samples, incr);

  947. 

  948.     offset = (offset - 32) & 511;

  949.     *synth_buf_offset = offset;

  950. }

  951. #endif

  952. 

  953. #define C3 FIXHR(0.86602540378443864676/2)

  954. 

  955. /* 0.5 / cos(pi*(2*i+1)/36) */

  956. static const INTFLOAT icos36[9] = {

  957.     FIXR(0.50190991877167369479),

  958.     FIXR(0.51763809020504152469), //0

  959.     FIXR(0.55168895948124587824),

  960.     FIXR(0.61038729438072803416),

  961.     FIXR(0.70710678118654752439), //1

  962.     FIXR(0.87172339781054900991),

  963.     FIXR(1.18310079157624925896),

  964.     FIXR(1.93185165257813657349), //2

  965.     FIXR(5.73685662283492756461),

  966. };

  967. 

  968. /* 0.5 / cos(pi*(2*i+1)/36) */

  969. static const INTFLOAT icos36h[9] = {

  970.     FIXHR(0.50190991877167369479/2),

  971.     FIXHR(0.51763809020504152469/2), //0

  972.     FIXHR(0.55168895948124587824/2),

  973.     FIXHR(0.61038729438072803416/2),

  974.     FIXHR(0.70710678118654752439/2), //1

  975.     FIXHR(0.87172339781054900991/2),

  976.     FIXHR(1.18310079157624925896/4),

  977.     FIXHR(1.93185165257813657349/4), //2

  978. //    FIXHR(5.73685662283492756461),

  979. };

  980. 

  981. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  982.    cases. */

  983. static void imdct12(INTFLOAT *out, INTFLOAT *in)

  984. {

  985.     INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;

  986. 

  987.     in0= in[0*3];

  988.     in1= in[1*3] + in[0*3];

  989.     in2= in[2*3] + in[1*3];

  990.     in3= in[3*3] + in[2*3];

  991.     in4= in[4*3] + in[3*3];

  992.     in5= in[5*3] + in[4*3];

  993.     in5 += in3;

  994.     in3 += in1;

  995. 

  996.     in2= MULH3(in2, C3, 2);

  997.     in3= MULH3(in3, C3, 4);

  998. 

  999.     t1 = in0 - in4;

  1000.     t2 = MULH3(in1 - in5, icos36h[4], 2);

  1001. 

  1002.     out[ 7]=

  1003.     out[10]= t1 + t2;

  1004.     out[ 1]=

  1005.     out[ 4]= t1 - t2;

  1006. 

  1007.     in0 += SHR(in4, 1);

  1008.     in4 = in0 + in2;

  1009.     in5 += 2*in1;

  1010.     in1 = MULH3(in5 + in3, icos36h[1], 1);

  1011.     out[ 8]=

  1012.     out[ 9]= in4 + in1;

  1013.     out[ 2]=

  1014.     out[ 3]= in4 - in1;

  1015. 

  1016.     in0 -= in2;

  1017.     in5 = MULH3(in5 - in3, icos36h[7], 2);

  1018.     out[ 0]=

  1019.     out[ 5]= in0 - in5;

  1020.     out[ 6]=

  1021.     out[11]= in0 + in5;

  1022. }

  1023. 

  1024. /* cos(pi*i/18) */

  1025. #define C1 FIXHR(0.98480775301220805936/2)

  1026. #define C2 FIXHR(0.93969262078590838405/2)

  1027. #define C3 FIXHR(0.86602540378443864676/2)

  1028. #define C4 FIXHR(0.76604444311897803520/2)

  1029. #define C5 FIXHR(0.64278760968653932632/2)

  1030. #define C6 FIXHR(0.5/2)

  1031. #define C7 FIXHR(0.34202014332566873304/2)

  1032. #define C8 FIXHR(0.17364817766693034885/2)

  1033. 

  1034. 

  1035. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1036. static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)

  1037. {

  1038.     int i, j;

  1039.     INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;

  1040.     INTFLOAT tmp[18], *tmp1, *in1;

  1041. 

  1042.     for(i=17;i>=1;i--)

  1043.         in[i] += in[i-1];

  1044.     for(i=17;i>=3;i-=2)

  1045.         in[i] += in[i-2];

  1046. 

  1047.     for(j=0;j<2;j++) {

  1048.         tmp1 = tmp + j;

  1049.         in1 = in + j;

  1050. 

  1051.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1052. 

  1053.         t3 = in1[2*0] + SHR(in1[2*6],1);

  1054.         t1 = in1[2*0] - in1[2*6];

  1055.         tmp1[ 6] = t1 - SHR(t2,1);

  1056.         tmp1[16] = t1 + t2;

  1057. 

  1058.         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);

  1059.         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);

  1060.         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);

  1061. 

  1062.         tmp1[10] = t3 - t0 - t2;

  1063.         tmp1[ 2] = t3 + t0 + t1;

  1064.         tmp1[14] = t3 + t2 - t1;

  1065. 

  1066.         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);

  1067.         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);

  1068.         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);

  1069.         t0 = MULH3(in1[2*3], C3, 2);

  1070. 

  1071.         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);

  1072. 

  1073.         tmp1[ 0] = t2 + t3 + t0;

  1074.         tmp1[12] = t2 + t1 - t0;

  1075.         tmp1[ 8] = t3 - t1 - t0;

  1076.     }

  1077. 

  1078.     i = 0;

  1079.     for(j=0;j<4;j++) {

  1080.         t0 = tmp[i];

  1081.         t1 = tmp[i + 2];

  1082.         s0 = t1 + t0;

  1083.         s2 = t1 - t0;

  1084. 

  1085.         t2 = tmp[i + 1];

  1086.         t3 = tmp[i + 3];

  1087.         s1 = MULH3(t3 + t2, icos36h[j], 2);

  1088.         s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);

  1089. 

  1090.         t0 = s0 + s1;

  1091.         t1 = s0 - s1;

  1092.         out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];

  1093.         out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];

  1094.         buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);

  1095.         buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);

  1096. 

  1097.         t0 = s2 + s3;

  1098.         t1 = s2 - s3;

  1099.         out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];

  1100.         out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];

  1101.         buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);

  1102.         buf[      + j] = MULH3(t0, win[18         + j], 1);

  1103.         i += 4;

  1104.     }

  1105. 

  1106.     s0 = tmp[16];

  1107.     s1 = MULH3(tmp[17], icos36h[4], 2);

  1108.     t0 = s0 + s1;

  1109.     t1 = s0 - s1;

  1110.     out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];

  1111.     out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];

  1112.     buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);

  1113.     buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);

  1114. }

  1115. 

  1116. /* return the number of decoded frames */

  1117. static int mp_decode_layer1(MPADecodeContext *s)

  1118. {

  1119.     int bound, i, v, n, ch, j, mant;

  1120.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1121.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1122. 

  1123.     if (s->mode == MPA_JSTEREO)

  1124.         bound = (s->mode_ext + 1) * 4;

  1125.     else

  1126.         bound = SBLIMIT;

  1127. 

  1128.     /* allocation bits */

  1129.     for(i=0;i<bound;i++) {

  1130.         for(ch=0;ch<s->nb_channels;ch++) {

  1131.             allocation[ch][i] = get_bits(&s->gb, 4);

  1132.         }

  1133.     }

  1134.     for(i=bound;i<SBLIMIT;i++) {

  1135.         allocation[0][i] = get_bits(&s->gb, 4);

  1136.     }

  1137. 

  1138.     /* scale factors */

  1139.     for(i=0;i<bound;i++) {

  1140.         for(ch=0;ch<s->nb_channels;ch++) {

  1141.             if (allocation[ch][i])

  1142.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1143.         }

  1144.     }

  1145.     for(i=bound;i<SBLIMIT;i++) {

  1146.         if (allocation[0][i]) {

  1147.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1148.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1149.         }

  1150.     }

  1151. 

  1152.     /* compute samples */

  1153.     for(j=0;j<12;j++) {

  1154.         for(i=0;i<bound;i++) {

  1155.             for(ch=0;ch<s->nb_channels;ch++) {

  1156.                 n = allocation[ch][i];

  1157.                 if (n) {

  1158.                     mant = get_bits(&s->gb, n + 1);

  1159.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1160.                 } else {

  1161.                     v = 0;

  1162.                 }

  1163.                 s->sb_samples[ch][j][i] = v;

  1164.             }

  1165.         }

  1166.         for(i=bound;i<SBLIMIT;i++) {

  1167.             n = allocation[0][i];

  1168.             if (n) {

  1169.                 mant = get_bits(&s->gb, n + 1);

  1170.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1171.                 s->sb_samples[0][j][i] = v;

  1172.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1173.                 s->sb_samples[1][j][i] = v;

  1174.             } else {

  1175.                 s->sb_samples[0][j][i] = 0;

  1176.                 s->sb_samples[1][j][i] = 0;

  1177.             }

  1178.         }

  1179.     }

  1180.     return 12;

  1181. }

  1182. 

  1183. static int mp_decode_layer2(MPADecodeContext *s)

  1184. {

  1185.     int sblimit; /* number of used subbands */

  1186.     const unsigned char *alloc_table;

  1187.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1188.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1189.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1190.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1191.     int scale, qindex, bits, steps, k, l, m, b;

  1192. 

  1193.     /* select decoding table */

  1194.     table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,

  1195.                             s->sample_rate, s->lsf);

  1196.     sblimit = ff_mpa_sblimit_table[table];

  1197.     alloc_table = ff_mpa_alloc_tables[table];

  1198. 

  1199.     if (s->mode == MPA_JSTEREO)

  1200.         bound = (s->mode_ext + 1) * 4;

  1201.     else

  1202.         bound = sblimit;

  1203. 

  1204.     dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

  1205. 

  1206.     /* sanity check */

  1207.     if( bound > sblimit ) bound = sblimit;

  1208. 

  1209.     /* parse bit allocation */

  1210.     j = 0;

  1211.     for(i=0;i<bound;i++) {

  1212.         bit_alloc_bits = alloc_table[j];

  1213.         for(ch=0;ch<s->nb_channels;ch++) {

  1214.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1215.         }

  1216.         j += 1 << bit_alloc_bits;

  1217.     }

  1218.     for(i=bound;i<sblimit;i++) {

  1219.         bit_alloc_bits = alloc_table[j];

  1220.         v = get_bits(&s->gb, bit_alloc_bits);

  1221.         bit_alloc[0][i] = v;

  1222.         bit_alloc[1][i] = v;

  1223.         j += 1 << bit_alloc_bits;

  1224.     }

  1225. 

  1226.     /* scale codes */

  1227.     for(i=0;i<sblimit;i++) {

  1228.         for(ch=0;ch<s->nb_channels;ch++) {

  1229.             if (bit_alloc[ch][i])

  1230.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1231.         }

  1232.     }

  1233. 

  1234.     /* scale factors */

  1235.     for(i=0;i<sblimit;i++) {

  1236.         for(ch=0;ch<s->nb_channels;ch++) {

  1237.             if (bit_alloc[ch][i]) {

  1238.                 sf = scale_factors[ch][i];

  1239.                 switch(scale_code[ch][i]) {

  1240.                 default:

  1241.                 case 0:

  1242.                     sf[0] = get_bits(&s->gb, 6);

  1243.                     sf[1] = get_bits(&s->gb, 6);

  1244.                     sf[2] = get_bits(&s->gb, 6);

  1245.                     break;

  1246.                 case 2:

  1247.                     sf[0] = get_bits(&s->gb, 6);

  1248.                     sf[1] = sf[0];

  1249.                     sf[2] = sf[0];

  1250.                     break;

  1251.                 case 1:

  1252.                     sf[0] = get_bits(&s->gb, 6);

  1253.                     sf[2] = get_bits(&s->gb, 6);

  1254.                     sf[1] = sf[0];

  1255.                     break;

  1256.                 case 3:

  1257.                     sf[0] = get_bits(&s->gb, 6);

  1258.                     sf[2] = get_bits(&s->gb, 6);

  1259.                     sf[1] = sf[2];

  1260.                     break;

  1261.                 }

  1262.             }

  1263.         }

  1264.     }

  1265. 

  1266.     /* samples */

  1267.     for(k=0;k<3;k++) {

  1268.         for(l=0;l<12;l+=3) {

  1269.             j = 0;

  1270.             for(i=0;i<bound;i++) {

  1271.                 bit_alloc_bits = alloc_table[j];

  1272.                 for(ch=0;ch<s->nb_channels;ch++) {

  1273.                     b = bit_alloc[ch][i];

  1274.                     if (b) {

  1275.                         scale = scale_factors[ch][i][k];

  1276.                         qindex = alloc_table[j+b];

  1277.                         bits = ff_mpa_quant_bits[qindex];

  1278.                         if (bits < 0) {

  1279.                             /* 3 values at the same time */

  1280.                             v = get_bits(&s->gb, -bits);

  1281.                             steps = ff_mpa_quant_steps[qindex];

  1282.                             s->sb_samples[ch][k * 12 + l + 0][i] =

  1283.                                 l2_unscale_group(steps, v % steps, scale);

  1284.                             v = v / steps;

  1285.                             s->sb_samples[ch][k * 12 + l + 1][i] =

  1286.                                 l2_unscale_group(steps, v % steps, scale);

  1287.                             v = v / steps;

  1288.                             s->sb_samples[ch][k * 12 + l + 2][i] =

  1289.                                 l2_unscale_group(steps, v, scale);

  1290.                         } else {

  1291.                             for(m=0;m<3;m++) {

  1292.                                 v = get_bits(&s->gb, bits);

  1293.                                 v = l1_unscale(bits - 1, v, scale);

  1294.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1295.                             }

  1296.                         }

  1297.                     } else {

  1298.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1299.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1300.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1301.                     }

  1302.                 }

  1303.                 /* next subband in alloc table */

  1304.                 j += 1 << bit_alloc_bits;

  1305.             }

  1306.             /* XXX: find a way to avoid this duplication of code */

  1307.             for(i=bound;i<sblimit;i++) {

  1308.                 bit_alloc_bits = alloc_table[j];

  1309.                 b = bit_alloc[0][i];

  1310.                 if (b) {

  1311.                     int mant, scale0, scale1;

  1312.                     scale0 = scale_factors[0][i][k];

  1313.                     scale1 = scale_factors[1][i][k];

  1314.                     qindex = alloc_table[j+b];

  1315.                     bits = ff_mpa_quant_bits[qindex];

  1316.                     if (bits < 0) {

  1317.                         /* 3 values at the same time */

  1318.                         v = get_bits(&s->gb, -bits);

  1319.                         steps = ff_mpa_quant_steps[qindex];

  1320.                         mant = v % steps;

  1321.                         v = v / steps;

  1322.                         s->sb_samples[0][k * 12 + l + 0][i] =

  1323.                             l2_unscale_group(steps, mant, scale0);

  1324.                         s->sb_samples[1][k * 12 + l + 0][i] =

  1325.                             l2_unscale_group(steps, mant, scale1);

  1326.                         mant = v % steps;

  1327.                         v = v / steps;

  1328.                         s->sb_samples[0][k * 12 + l + 1][i] =

  1329.                             l2_unscale_group(steps, mant, scale0);

  1330.                         s->sb_samples[1][k * 12 + l + 1][i] =

  1331.                             l2_unscale_group(steps, mant, scale1);

  1332.                         s->sb_samples[0][k * 12 + l + 2][i] =

  1333.                             l2_unscale_group(steps, v, scale0);

  1334.                         s->sb_samples[1][k * 12 + l + 2][i] =

  1335.                             l2_unscale_group(steps, v, scale1);

  1336.                     } else {

  1337.                         for(m=0;m<3;m++) {

  1338.                             mant = get_bits(&s->gb, bits);

  1339.                             s->sb_samples[0][k * 12 + l + m][i] =

  1340.                                 l1_unscale(bits - 1, mant, scale0);

  1341.                             s->sb_samples[1][k * 12 + l + m][i] =

  1342.                                 l1_unscale(bits - 1, mant, scale1);

  1343.                         }

  1344.                     }

  1345.                 } else {

  1346.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1347.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1348.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1349.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1350.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1351.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1352.                 }

  1353.                 /* next subband in alloc table */

  1354.                 j += 1 << bit_alloc_bits;

  1355.             }

  1356.             /* fill remaining samples to zero */

  1357.             for(i=sblimit;i<SBLIMIT;i++) {

  1358.                 for(ch=0;ch<s->nb_channels;ch++) {

  1359.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1360.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1361.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1362.                 }

  1363.             }

  1364.         }

  1365.     }

  1366.     return 3 * 12;

  1367. }

  1368. 

  1369. #define SPLIT(dst,sf,n)\

  1370.     if(n==3){\

  1371.         int m= (sf*171)>>9;\

  1372.         dst= sf - 3*m;\

  1373.         sf=m;\

  1374.     }else if(n==4){\

  1375.         dst= sf&3;\

  1376.         sf>>=2;\

  1377.     }else if(n==5){\

  1378.         int m= (sf*205)>>10;\

  1379.         dst= sf - 5*m;\

  1380.         sf=m;\

  1381.     }else if(n==6){\

  1382.         int m= (sf*171)>>10;\

  1383.         dst= sf - 6*m;\

  1384.         sf=m;\

  1385.     }else{\

  1386.         dst=0;\

  1387.     }

  1388. 

  1389. static av_always_inline void lsf_sf_expand(int *slen,

  1390.                                  int sf, int n1, int n2, int n3)

  1391. {

  1392.     SPLIT(slen[3], sf, n3)

  1393.     SPLIT(slen[2], sf, n2)

  1394.     SPLIT(slen[1], sf, n1)

  1395.     slen[0] = sf;

  1396. }

  1397. 

  1398. static void exponents_from_scale_factors(MPADecodeContext *s,

  1399.                                          GranuleDef *g,

  1400.                                          int16_t *exponents)

  1401. {

  1402.     const uint8_t *bstab, *pretab;

  1403.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1404.     int16_t *exp_ptr;

  1405. 

  1406.     exp_ptr = exponents;

  1407.     gain = g->global_gain - 210;

  1408.     shift = g->scalefac_scale + 1;

  1409. 

  1410.     bstab = band_size_long[s->sample_rate_index];

  1411.     pretab = mpa_pretab[g->preflag];

  1412.     for(i=0;i<g->long_end;i++) {

  1413.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;

  1414.         len = bstab[i];

  1415.         for(j=len;j>0;j--)

  1416.             *exp_ptr++ = v0;

  1417.     }

  1418. 

  1419.     if (g->short_start < 13) {

  1420.         bstab = band_size_short[s->sample_rate_index];

  1421.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1422.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1423.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1424.         k = g->long_end;

  1425.         for(i=g->short_start;i<13;i++) {

  1426.             len = bstab[i];

  1427.             for(l=0;l<3;l++) {

  1428.                 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;

  1429.                 for(j=len;j>0;j--)

  1430.                 *exp_ptr++ = v0;

  1431.             }

  1432.         }

  1433.     }

  1434. }

  1435. 

  1436. /* handle n = 0 too */

  1437. static inline int get_bitsz(GetBitContext *s, int n)

  1438. {

  1439.     if (n == 0)

  1440.         return 0;

  1441.     else

  1442.         return get_bits(s, n);

  1443. }

  1444. 

  1445. 

  1446. static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){

  1447.     if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){

  1448.         s->gb= s->in_gb;

  1449.         s->in_gb.buffer=NULL;

  1450.         assert((get_bits_count(&s->gb) & 7) == 0);

  1451.         skip_bits_long(&s->gb, *pos - *end_pos);

  1452.         *end_pos2=

  1453.         *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;

  1454.         *pos= get_bits_count(&s->gb);

  1455.     }

  1456. }

  1457. 

  1458. /* Following is a optimized code for

  1459.             INTFLOAT v = *src

  1460.             if(get_bits1(&s->gb))

  1461.                 v = -v;

  1462.             *dst = v;

  1463. */

  1464. #if CONFIG_FLOAT

  1465. #define READ_FLIP_SIGN(dst,src)\

  1466.             v = AV_RN32A(src) ^ (get_bits1(&s->gb)<<31);\

  1467.             AV_WN32A(dst, v);

  1468. #else

  1469. #define READ_FLIP_SIGN(dst,src)\

  1470.             v= -get_bits1(&s->gb);\

  1471.             *(dst) = (*(src) ^ v) - v;

  1472. #endif

  1473. 

  1474. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1475.                           int16_t *exponents, int end_pos2)

  1476. {

  1477.     int s_index;

  1478.     int i;

  1479.     int last_pos, bits_left;

  1480.     VLC *vlc;

  1481.     int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);

  1482. 

  1483.     /* low frequencies (called big values) */

  1484.     s_index = 0;

  1485.     for(i=0;i<3;i++) {

  1486.         int j, k, l, linbits;

  1487.         j = g->region_size[i];

  1488.         if (j == 0)

  1489.             continue;

  1490.         /* select vlc table */

  1491.         k = g->table_select[i];

  1492.         l = mpa_huff_data[k][0];

  1493.         linbits = mpa_huff_data[k][1];

  1494.         vlc = &huff_vlc[l];

  1495. 

  1496.         if(!l){

  1497.             memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);

  1498.             s_index += 2*j;

  1499.             continue;

  1500.         }

  1501. 

  1502.         /* read huffcode and compute each couple */

  1503.         for(;j>0;j--) {

  1504.             int exponent, x, y;

  1505.             int v;

  1506.             int pos= get_bits_count(&s->gb);

  1507. 

  1508.             if (pos >= end_pos){

  1509. //                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1510.                 switch_buffer(s, &pos, &end_pos, &end_pos2);

  1511. //                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);

  1512.                 if(pos >= end_pos)

  1513.                     break;

  1514.             }

  1515.             y = get_vlc2(&s->gb, vlc->table, 7, 3);

  1516. 

  1517.             if(!y){

  1518.                 g->sb_hybrid[s_index  ] =

  1519.                 g->sb_hybrid[s_index+1] = 0;

  1520.                 s_index += 2;

  1521.                 continue;

  1522.             }

  1523. 

  1524.             exponent= exponents[s_index];

  1525. 

  1526.             dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",

  1527.                     i, g->region_size[i] - j, x, y, exponent);

  1528.             if(y&16){

  1529.                 x = y >> 5;

  1530.                 y = y & 0x0f;

  1531.                 if (x < 15){

  1532.                     READ_FLIP_SIGN(g->sb_hybrid+s_index, RENAME(expval_table)[ exponent ]+x)

  1533.                 }else{

  1534.                     x += get_bitsz(&s->gb, linbits);

  1535.                     v = l3_unscale(x, exponent);

  1536.                     if (get_bits1(&s->gb))

  1537.                         v = -v;

  1538.                     g->sb_hybrid[s_index] = v;

  1539.                 }

  1540.                 if (y < 15){

  1541.                     READ_FLIP_SIGN(g->sb_hybrid+s_index+1, RENAME(expval_table)[ exponent ]+y)

  1542.                 }else{

  1543.                     y += get_bitsz(&s->gb, linbits);

  1544.                     v = l3_unscale(y, exponent);

  1545.                     if (get_bits1(&s->gb))

  1546.                         v = -v;

  1547.                     g->sb_hybrid[s_index+1] = v;

  1548.                 }

  1549.             }else{

  1550.                 x = y >> 5;

  1551.                 y = y & 0x0f;

  1552.                 x += y;

  1553.                 if (x < 15){

  1554.                     READ_FLIP_SIGN(g->sb_hybrid+s_index+!!y, RENAME(expval_table)[ exponent ]+x)

  1555.                 }else{

  1556.                     x += get_bitsz(&s->gb, linbits);

  1557.                     v = l3_unscale(x, exponent);

  1558.                     if (get_bits1(&s->gb))

  1559.                         v = -v;

  1560.                     g->sb_hybrid[s_index+!!y] = v;

  1561.                 }

  1562.                 g->sb_hybrid[s_index+ !y] = 0;

  1563.             }

  1564.             s_index+=2;

  1565.         }

  1566.     }

  1567. 

  1568.     /* high frequencies */

  1569.     vlc = &huff_quad_vlc[g->count1table_select];

  1570.     last_pos=0;

  1571.     while (s_index <= 572) {

  1572.         int pos, code;

  1573.         pos = get_bits_count(&s->gb);

  1574.         if (pos >= end_pos) {

  1575.             if (pos > end_pos2 && last_pos){

  1576.                 /* some encoders generate an incorrect size for this

  1577.                    part. We must go back into the data */

  1578.                 s_index -= 4;

  1579.                 skip_bits_long(&s->gb, last_pos - pos);

  1580.                 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);

  1581.                 if(s->error_recognition >= FF_ER_COMPLIANT)

  1582.                     s_index=0;

  1583.                 break;

  1584.             }

  1585. //                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1586.             switch_buffer(s, &pos, &end_pos, &end_pos2);

  1587. //                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);

  1588.             if(pos >= end_pos)

  1589.                 break;

  1590.         }

  1591.         last_pos= pos;

  1592. 

  1593.         code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);

  1594.         dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);

  1595.         g->sb_hybrid[s_index+0]=

  1596.         g->sb_hybrid[s_index+1]=

  1597.         g->sb_hybrid[s_index+2]=

  1598.         g->sb_hybrid[s_index+3]= 0;

  1599.         while(code){

  1600.             static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};

  1601.             int v;

  1602.             int pos= s_index+idxtab[code];

  1603.             code ^= 8>>idxtab[code];

  1604.             READ_FLIP_SIGN(g->sb_hybrid+pos, RENAME(exp_table)+exponents[pos])

  1605.         }

  1606.         s_index+=4;

  1607.     }

  1608.     /* skip extension bits */

  1609.     bits_left = end_pos2 - get_bits_count(&s->gb);

  1610. //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);

  1611.     if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {

  1612.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1613.         s_index=0;

  1614.     }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){

  1615.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1616.         s_index=0;

  1617.     }

  1618.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));

  1619.     skip_bits_long(&s->gb, bits_left);

  1620. 

  1621.     i= get_bits_count(&s->gb);

  1622.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1623. 

  1624.     return 0;

  1625. }

  1626. 

  1627. /* Reorder short blocks from bitstream order to interleaved order. It

  1628.    would be faster to do it in parsing, but the code would be far more

  1629.    complicated */

  1630. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1631. {

  1632.     int i, j, len;

  1633.     INTFLOAT *ptr, *dst, *ptr1;

  1634.     INTFLOAT tmp[576];

  1635. 

  1636.     if (g->block_type != 2)

  1637.         return;

  1638. 

  1639.     if (g->switch_point) {

  1640.         if (s->sample_rate_index != 8) {

  1641.             ptr = g->sb_hybrid + 36;

  1642.         } else {

  1643.             ptr = g->sb_hybrid + 48;

  1644.         }

  1645.     } else {

  1646.         ptr = g->sb_hybrid;

  1647.     }

  1648. 

  1649.     for(i=g->short_start;i<13;i++) {

  1650.         len = band_size_short[s->sample_rate_index][i];

  1651.         ptr1 = ptr;

  1652.         dst = tmp;

  1653.         for(j=len;j>0;j--) {

  1654.             *dst++ = ptr[0*len];

  1655.             *dst++ = ptr[1*len];

  1656.             *dst++ = ptr[2*len];

  1657.             ptr++;

  1658.         }

  1659.         ptr+=2*len;

  1660.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1661.     }

  1662. }

  1663. 

  1664. #define ISQRT2 FIXR(0.70710678118654752440)

  1665. 

  1666. static void compute_stereo(MPADecodeContext *s,

  1667.                            GranuleDef *g0, GranuleDef *g1)

  1668. {

  1669.     int i, j, k, l;

  1670.     int sf_max, sf, len, non_zero_found;

  1671.     INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;

  1672.     int non_zero_found_short[3];

  1673. 

  1674.     /* intensity stereo */

  1675.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1676.         if (!s->lsf) {

  1677.             is_tab = is_table;

  1678.             sf_max = 7;

  1679.         } else {

  1680.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1681.             sf_max = 16;

  1682.         }

  1683. 

  1684.         tab0 = g0->sb_hybrid + 576;

  1685.         tab1 = g1->sb_hybrid + 576;

  1686. 

  1687.         non_zero_found_short[0] = 0;

  1688.         non_zero_found_short[1] = 0;

  1689.         non_zero_found_short[2] = 0;

  1690.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1691.         for(i = 12;i >= g1->short_start;i--) {

  1692.             /* for last band, use previous scale factor */

  1693.             if (i != 11)

  1694.                 k -= 3;

  1695.             len = band_size_short[s->sample_rate_index][i];

  1696.             for(l=2;l>=0;l--) {

  1697.                 tab0 -= len;

  1698.                 tab1 -= len;

  1699.                 if (!non_zero_found_short[l]) {

  1700.                     /* test if non zero band. if so, stop doing i-stereo */

  1701.                     for(j=0;j<len;j++) {

  1702.                         if (tab1[j] != 0) {

  1703.                             non_zero_found_short[l] = 1;

  1704.                             goto found1;

  1705.                         }

  1706.                     }

  1707.                     sf = g1->scale_factors[k + l];

  1708.                     if (sf >= sf_max)

  1709.                         goto found1;

  1710. 

  1711.                     v1 = is_tab[0][sf];

  1712.                     v2 = is_tab[1][sf];

  1713.                     for(j=0;j<len;j++) {

  1714.                         tmp0 = tab0[j];

  1715.                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1716.                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1717.                     }

  1718.                 } else {

  1719.                 found1:

  1720.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1721.                         /* lower part of the spectrum : do ms stereo

  1722.                            if enabled */

  1723.                         for(j=0;j<len;j++) {

  1724.                             tmp0 = tab0[j];

  1725.                             tmp1 = tab1[j];

  1726.                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1727.                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1728.                         }

  1729.                     }

  1730.                 }

  1731.             }

  1732.         }

  1733. 

  1734.         non_zero_found = non_zero_found_short[0] |

  1735.             non_zero_found_short[1] |

  1736.             non_zero_found_short[2];

  1737. 

  1738.         for(i = g1->long_end - 1;i >= 0;i--) {

  1739.             len = band_size_long[s->sample_rate_index][i];

  1740.             tab0 -= len;

  1741.             tab1 -= len;

  1742.             /* test if non zero band. if so, stop doing i-stereo */

  1743.             if (!non_zero_found) {

  1744.                 for(j=0;j<len;j++) {

  1745.                     if (tab1[j] != 0) {

  1746.                         non_zero_found = 1;

  1747.                         goto found2;

  1748.                     }

  1749.                 }

  1750.                 /* for last band, use previous scale factor */

  1751.                 k = (i == 21) ? 20 : i;

  1752.                 sf = g1->scale_factors[k];

  1753.                 if (sf >= sf_max)

  1754.                     goto found2;

  1755.                 v1 = is_tab[0][sf];

  1756.                 v2 = is_tab[1][sf];

  1757.                 for(j=0;j<len;j++) {

  1758.                     tmp0 = tab0[j];

  1759.                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1760.                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1761.                 }

  1762.             } else {

  1763.             found2:

  1764.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1765.                     /* lower part of the spectrum : do ms stereo

  1766.                        if enabled */

  1767.                     for(j=0;j<len;j++) {

  1768.                         tmp0 = tab0[j];

  1769.                         tmp1 = tab1[j];

  1770.                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1771.                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1772.                     }

  1773.                 }

  1774.             }

  1775.         }

  1776.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1777.         /* ms stereo ONLY */

  1778.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1779.            global gain */

  1780.         tab0 = g0->sb_hybrid;

  1781.         tab1 = g1->sb_hybrid;

  1782.         for(i=0;i<576;i++) {

  1783.             tmp0 = tab0[i];

  1784.             tmp1 = tab1[i];

  1785.             tab0[i] = tmp0 + tmp1;

  1786.             tab1[i] = tmp0 - tmp1;

  1787.         }

  1788.     }

  1789. }

  1790. 

  1791. static void compute_antialias_integer(MPADecodeContext *s,

  1792.                               GranuleDef *g)

  1793. {

  1794.     int32_t *ptr, *csa;

  1795.     int n, i;

  1796. 

  1797.     /* we antialias only "long" bands */

  1798.     if (g->block_type == 2) {

  1799.         if (!g->switch_point)

  1800.             return;

  1801.         /* XXX: check this for 8000Hz case */

  1802.         n = 1;

  1803.     } else {

  1804.         n = SBLIMIT - 1;

  1805.     }

  1806. 

  1807.     ptr = g->sb_hybrid + 18;

  1808.     for(i = n;i > 0;i--) {

  1809.         int tmp0, tmp1, tmp2;

  1810.         csa = &csa_table[0][0];

  1811. #define INT_AA(j) \

  1812.             tmp0 = ptr[-1-j];\

  1813.             tmp1 = ptr[   j];\

  1814.             tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\

  1815.             ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\

  1816.             ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));

  1817. 

  1818.         INT_AA(0)

  1819.         INT_AA(1)

  1820.         INT_AA(2)

  1821.         INT_AA(3)

  1822.         INT_AA(4)

  1823.         INT_AA(5)

  1824.         INT_AA(6)

  1825.         INT_AA(7)

  1826. 

  1827.         ptr += 18;

  1828.     }

  1829. }

  1830. 

  1831. static void compute_antialias_float(MPADecodeContext *s,

  1832.                               GranuleDef *g)

  1833. {

  1834.     float *ptr;

  1835.     int n, i;

  1836. 

  1837.     /* we antialias only "long" bands */

  1838.     if (g->block_type == 2) {

  1839.         if (!g->switch_point)

  1840.             return;

  1841.         /* XXX: check this for 8000Hz case */

  1842.         n = 1;

  1843.     } else {

  1844.         n = SBLIMIT - 1;

  1845.     }

  1846. 

  1847.     ptr = g->sb_hybrid + 18;

  1848.     for(i = n;i > 0;i--) {

  1849.         float tmp0, tmp1;

  1850.         float *csa = &csa_table_float[0][0];

  1851. #define FLOAT_AA(j)\

  1852.         tmp0= ptr[-1-j];\

  1853.         tmp1= ptr[   j];\

  1854.         ptr[-1-j] = tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j];\

  1855.         ptr[   j] = tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j];

  1856. 

  1857.         FLOAT_AA(0)

  1858.         FLOAT_AA(1)

  1859.         FLOAT_AA(2)

  1860.         FLOAT_AA(3)

  1861.         FLOAT_AA(4)

  1862.         FLOAT_AA(5)

  1863.         FLOAT_AA(6)

  1864.         FLOAT_AA(7)

  1865. 

  1866.         ptr += 18;

  1867.     }

  1868. }

  1869. 

  1870. static void compute_imdct(MPADecodeContext *s,

  1871.                           GranuleDef *g,

  1872.                           INTFLOAT *sb_samples,

  1873.                           INTFLOAT *mdct_buf)

  1874. {

  1875.     INTFLOAT *win, *win1, *out_ptr, *ptr, *buf, *ptr1;

  1876.     INTFLOAT out2[12];

  1877.     int i, j, mdct_long_end, sblimit;

  1878. 

  1879.     /* find last non zero block */

  1880.     ptr = g->sb_hybrid + 576;

  1881.     ptr1 = g->sb_hybrid + 2 * 18;

  1882.     while (ptr >= ptr1) {

  1883.         int32_t *p;

  1884.         ptr -= 6;

  1885.         p= (int32_t*)ptr;

  1886.         if(p[0] | p[1] | p[2] | p[3] | p[4] | p[5])

  1887.             break;

  1888.     }

  1889.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1890. 

  1891.     if (g->block_type == 2) {

  1892.         /* XXX: check for 8000 Hz */

  1893.         if (g->switch_point)

  1894.             mdct_long_end = 2;

  1895.         else

  1896.             mdct_long_end = 0;

  1897.     } else {

  1898.         mdct_long_end = sblimit;

  1899.     }

  1900. 

  1901.     buf = mdct_buf;

  1902.     ptr = g->sb_hybrid;

  1903.     for(j=0;j<mdct_long_end;j++) {

  1904.         /* apply window & overlap with previous buffer */

  1905.         out_ptr = sb_samples + j;

  1906.         /* select window */

  1907.         if (g->switch_point && j < 2)

  1908.             win1 = mdct_win[0];

  1909.         else

  1910.             win1 = mdct_win[g->block_type];

  1911.         /* select frequency inversion */

  1912.         win = win1 + ((4 * 36) & -(j & 1));

  1913.         imdct36(out_ptr, buf, ptr, win);

  1914.         out_ptr += 18*SBLIMIT;

  1915.         ptr += 18;

  1916.         buf += 18;

  1917.     }

  1918.     for(j=mdct_long_end;j<sblimit;j++) {

  1919.         /* select frequency inversion */

  1920.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  1921.         out_ptr = sb_samples + j;

  1922. 

  1923.         for(i=0; i<6; i++){

  1924.             *out_ptr = buf[i];

  1925.             out_ptr += SBLIMIT;

  1926.         }

  1927.         imdct12(out2, ptr + 0);

  1928.         for(i=0;i<6;i++) {

  1929.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*1];

  1930.             buf[i + 6*2] = MULH3(out2[i + 6], win[i + 6], 1);

  1931.             out_ptr += SBLIMIT;

  1932.         }

  1933.         imdct12(out2, ptr + 1);

  1934.         for(i=0;i<6;i++) {

  1935.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*2];

  1936.             buf[i + 6*0] = MULH3(out2[i + 6], win[i + 6], 1);

  1937.             out_ptr += SBLIMIT;

  1938.         }

  1939.         imdct12(out2, ptr + 2);

  1940.         for(i=0;i<6;i++) {

  1941.             buf[i + 6*0] = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*0];

  1942.             buf[i + 6*1] = MULH3(out2[i + 6], win[i + 6], 1);

  1943.             buf[i + 6*2] = 0;

  1944.         }

  1945.         ptr += 18;

  1946.         buf += 18;

  1947.     }

  1948.     /* zero bands */

  1949.     for(j=sblimit;j<SBLIMIT;j++) {

  1950.         /* overlap */

  1951.         out_ptr = sb_samples + j;

  1952.         for(i=0;i<18;i++) {

  1953.             *out_ptr = buf[i];

  1954.             buf[i] = 0;

  1955.             out_ptr += SBLIMIT;

  1956.         }

  1957.         buf += 18;

  1958.     }

  1959. }

  1960. 

  1961. /* main layer3 decoding function */

  1962. static int mp_decode_layer3(MPADecodeContext *s)

  1963. {

  1964.     int nb_granules, main_data_begin, private_bits;

  1965.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1966.     GranuleDef *g;

  1967.     int16_t exponents[576]; //FIXME try INTFLOAT

  1968. 

  1969.     /* read side info */

  1970.     if (s->lsf) {

  1971.         main_data_begin = get_bits(&s->gb, 8);

  1972.         private_bits = get_bits(&s->gb, s->nb_channels);

  1973.         nb_granules = 1;

  1974.     } else {

  1975.         main_data_begin = get_bits(&s->gb, 9);

  1976.         if (s->nb_channels == 2)

  1977.             private_bits = get_bits(&s->gb, 3);

  1978.         else

  1979.             private_bits = get_bits(&s->gb, 5);

  1980.         nb_granules = 2;

  1981.         for(ch=0;ch<s->nb_channels;ch++) {

  1982.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1983.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1984.         }

  1985.     }

  1986. 

  1987.     for(gr=0;gr<nb_granules;gr++) {

  1988.         for(ch=0;ch<s->nb_channels;ch++) {

  1989.             dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1990.             g = &s->granules[ch][gr];

  1991.             g->part2_3_length = get_bits(&s->gb, 12);

  1992.             g->big_values = get_bits(&s->gb, 9);

  1993.             if(g->big_values > 288){

  1994.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1995.                 return -1;

  1996.             }

  1997. 

  1998.             g->global_gain = get_bits(&s->gb, 8);

  1999.             /* if MS stereo only is selected, we precompute the

  2000.                1/sqrt(2) renormalization factor */

  2001.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  2002.                 MODE_EXT_MS_STEREO)

  2003.                 g->global_gain -= 2;

  2004.             if (s->lsf)

  2005.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2006.             else

  2007.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2008.             blocksplit_flag = get_bits1(&s->gb);

  2009.             if (blocksplit_flag) {

  2010.                 g->block_type = get_bits(&s->gb, 2);

  2011.                 if (g->block_type == 0){

  2012.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  2013.                     return -1;

  2014.                 }

  2015.                 g->switch_point = get_bits1(&s->gb);

  2016.                 for(i=0;i<2;i++)

  2017.                     g->table_select[i] = get_bits(&s->gb, 5);

  2018.                 for(i=0;i<3;i++)

  2019.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2020.                 ff_init_short_region(s, g);

  2021.             } else {

  2022.                 int region_address1, region_address2;

  2023.                 g->block_type = 0;

  2024.                 g->switch_point = 0;

  2025.                 for(i=0;i<3;i++)

  2026.                     g->table_select[i] = get_bits(&s->gb, 5);

  2027.                 /* compute huffman coded region sizes */

  2028.                 region_address1 = get_bits(&s->gb, 4);

  2029.                 region_address2 = get_bits(&s->gb, 3);

  2030.                 dprintf(s->avctx, "region1=%d region2=%d\n",

  2031.                         region_address1, region_address2);

  2032.                 ff_init_long_region(s, g, region_address1, region_address2);

  2033.             }

  2034.             ff_region_offset2size(g);

  2035.             ff_compute_band_indexes(s, g);

  2036. 

  2037.             g->preflag = 0;

  2038.             if (!s->lsf)

  2039.                 g->preflag = get_bits1(&s->gb);

  2040.             g->scalefac_scale = get_bits1(&s->gb);

  2041.             g->count1table_select = get_bits1(&s->gb);

  2042.             dprintf(s->avctx, "block_type=%d switch_point=%d\n",

  2043.                     g->block_type, g->switch_point);

  2044.         }

  2045.     }

  2046. 

  2047.   if (!s->adu_mode) {

  2048.     const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  2049.     assert((get_bits_count(&s->gb) & 7) == 0);

  2050.     /* now we get bits from the main_data_begin offset */

  2051.     dprintf(s->avctx, "seekback: %d\n", main_data_begin);

  2052. //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);

  2053. 

  2054.     memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);

  2055.     s->in_gb= s->gb;

  2056.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  2057.         skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));

  2058.   }

  2059. 

  2060.     for(gr=0;gr<nb_granules;gr++) {

  2061.         for(ch=0;ch<s->nb_channels;ch++) {

  2062.             g = &s->granules[ch][gr];

  2063.             if(get_bits_count(&s->gb)<0){

  2064.                 av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",

  2065.                                             main_data_begin, s->last_buf_size, gr);

  2066.                 skip_bits_long(&s->gb, g->part2_3_length);

  2067.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  2068.                 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){

  2069.                     skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);

  2070.                     s->gb= s->in_gb;

  2071.                     s->in_gb.buffer=NULL;

  2072.                 }

  2073.                 continue;

  2074.             }

  2075. 

  2076.             bits_pos = get_bits_count(&s->gb);

  2077. 

  2078.             if (!s->lsf) {

  2079.                 uint8_t *sc;

  2080.                 int slen, slen1, slen2;

  2081. 

  2082.                 /* MPEG1 scale factors */

  2083.                 slen1 = slen_table[0][g->scalefac_compress];

  2084.                 slen2 = slen_table[1][g->scalefac_compress];

  2085.                 dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  2086.                 if (g->block_type == 2) {

  2087.                     n = g->switch_point ? 17 : 18;

  2088.                     j = 0;

  2089.                     if(slen1){

  2090.                         for(i=0;i<n;i++)

  2091.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  2092.                     }else{

  2093.                         for(i=0;i<n;i++)

  2094.                             g->scale_factors[j++] = 0;

  2095.                     }

  2096.                     if(slen2){

  2097.                         for(i=0;i<18;i++)

  2098.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  2099.                         for(i=0;i<3;i++)

  2100.                             g->scale_factors[j++] = 0;

  2101.                     }else{

  2102.                         for(i=0;i<21;i++)

  2103.                             g->scale_factors[j++] = 0;

  2104.                     }

  2105.                 } else {

  2106.                     sc = s->granules[ch][0].scale_factors;

  2107.                     j = 0;

  2108.                     for(k=0;k<4;k++) {

  2109.                         n = (k == 0 ? 6 : 5);

  2110.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2111.                             slen = (k < 2) ? slen1 : slen2;

  2112.                             if(slen){

  2113.                                 for(i=0;i<n;i++)

  2114.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  2115.                             }else{

  2116.                                 for(i=0;i<n;i++)

  2117.                                     g->scale_factors[j++] = 0;

  2118.                             }

  2119.                         } else {

  2120.                             /* simply copy from last granule */

  2121.                             for(i=0;i<n;i++) {

  2122.                                 g->scale_factors[j] = sc[j];

  2123.                                 j++;

  2124.                             }

  2125.                         }

  2126.                     }

  2127.                     g->scale_factors[j++] = 0;

  2128.                 }

  2129.             } else {

  2130.                 int tindex, tindex2, slen[4], sl, sf;

  2131. 

  2132.                 /* LSF scale factors */

  2133.                 if (g->block_type == 2) {

  2134.                     tindex = g->switch_point ? 2 : 1;

  2135.                 } else {

  2136.                     tindex = 0;

  2137.                 }

  2138.                 sf = g->scalefac_compress;

  2139.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2140.                     /* intensity stereo case */

  2141.                     sf >>= 1;

  2142.                     if (sf < 180) {

  2143.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2144.                         tindex2 = 3;

  2145.                     } else if (sf < 244) {

  2146.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2147.                         tindex2 = 4;

  2148.                     } else {

  2149.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2150.                         tindex2 = 5;

  2151.                     }

  2152.                 } else {

  2153.                     /* normal case */

  2154.                     if (sf < 400) {

  2155.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2156.                         tindex2 = 0;

  2157.                     } else if (sf < 500) {

  2158.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2159.                         tindex2 = 1;

  2160.                     } else {

  2161.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2162.                         tindex2 = 2;

  2163.                         g->preflag = 1;

  2164.                     }

  2165.                 }

  2166. 

  2167.                 j = 0;

  2168.                 for(k=0;k<4;k++) {

  2169.                     n = lsf_nsf_table[tindex2][tindex][k];

  2170.                     sl = slen[k];

  2171.                     if(sl){

  2172.                         for(i=0;i<n;i++)

  2173.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  2174.                     }else{

  2175.                         for(i=0;i<n;i++)

  2176.                             g->scale_factors[j++] = 0;

  2177.                     }

  2178.                 }

  2179.                 /* XXX: should compute exact size */

  2180.                 for(;j<40;j++)

  2181.                     g->scale_factors[j] = 0;

  2182.             }

  2183. 

  2184.             exponents_from_scale_factors(s, g, exponents);

  2185. 

  2186.             /* read Huffman coded residue */

  2187.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  2188.         } /* ch */

  2189. 

  2190.         if (s->nb_channels == 2)

  2191.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  2192. 

  2193.         for(ch=0;ch<s->nb_channels;ch++) {

  2194.             g = &s->granules[ch][gr];

  2195. 

  2196.             reorder_block(s, g);

  2197.             compute_antialias(s, g);

  2198.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  2199.         }

  2200.     } /* gr */

  2201.     if(get_bits_count(&s->gb)<0)

  2202.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  2203.     return nb_granules * 18;

  2204. }

  2205. 

  2206. static int mp_decode_frame(MPADecodeContext *s,

  2207.                            OUT_INT *samples, const uint8_t *buf, int buf_size)

  2208. {

  2209.     int i, nb_frames, ch;

  2210.     OUT_INT *samples_ptr;

  2211. 

  2212.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);

  2213. 

  2214.     /* skip error protection field */

  2215.     if (s->error_protection)

  2216.         skip_bits(&s->gb, 16);

  2217. 

  2218.     dprintf(s->avctx, "frame %d:\n", s->frame_count);

  2219.     switch(s->layer) {

  2220.     case 1:

  2221.         s->avctx->frame_size = 384;

  2222.         nb_frames = mp_decode_layer1(s);

  2223.         break;

  2224.     case 2:

  2225.         s->avctx->frame_size = 1152;

  2226.         nb_frames = mp_decode_layer2(s);

  2227.         break;

  2228.     case 3:

  2229.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  2230.     default:

  2231.         nb_frames = mp_decode_layer3(s);

  2232. 

  2233.         s->last_buf_size=0;

  2234.         if(s->in_gb.buffer){

  2235.             align_get_bits(&s->gb);

  2236.             i= get_bits_left(&s->gb)>>3;

  2237.             if(i >= 0 && i <= BACKSTEP_SIZE){

  2238.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  2239.                 s->last_buf_size=i;

  2240.             }else

  2241.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  2242.             s->gb= s->in_gb;

  2243.             s->in_gb.buffer= NULL;

  2244.         }

  2245. 

  2246.         align_get_bits(&s->gb);

  2247.         assert((get_bits_count(&s->gb) & 7) == 0);

  2248.         i= get_bits_left(&s->gb)>>3;

  2249. 

  2250.         if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){

  2251.             if(i<0)

  2252.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  2253.             i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  2254.         }

  2255.         assert(i <= buf_size - HEADER_SIZE && i>= 0);

  2256.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  2257.         s->last_buf_size += i;

  2258. 

  2259.         break;

  2260.     }

  2261. 

  2262.     /* apply the synthesis filter */

  2263.     for(ch=0;ch<s->nb_channels;ch++) {

  2264.         samples_ptr = samples + ch;

  2265.         for(i=0;i<nb_frames;i++) {

  2266.             RENAME(ff_mpa_synth_filter)(

  2267. #if CONFIG_FLOAT

  2268.                          s,

  2269. #endif

  2270.                          s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2271.                          RENAME(ff_mpa_synth_window), &s->dither_state,

  2272.                          samples_ptr, s->nb_channels,

  2273.                          s->sb_samples[ch][i]);

  2274.             samples_ptr += 32 * s->nb_channels;

  2275.         }

  2276.     }

  2277. 

  2278.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2279. }

  2280. 

  2281. static int decode_frame(AVCodecContext * avctx,

  2282.                         void *data, int *data_size,

  2283.                         AVPacket *avpkt)

  2284. {

  2285.     const uint8_t *buf = avpkt->data;

  2286.     int buf_size = avpkt->size;

  2287.     MPADecodeContext *s = avctx->priv_data;

  2288.     uint32_t header;

  2289.     int out_size;

  2290.     OUT_INT *out_samples = data;

  2291. 

  2292.     if(buf_size < HEADER_SIZE)

  2293.         return -1;

  2294. 

  2295.     header = AV_RB32(buf);

  2296.     if(ff_mpa_check_header(header) < 0){

  2297.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  2298.         return -1;

  2299.     }

  2300. 

  2301.     if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {

  2302.         /* free format: prepare to compute frame size */

  2303.         s->frame_size = -1;

  2304.         return -1;

  2305.     }

  2306.     /* update codec info */

  2307.     avctx->channels = s->nb_channels;

  2308.     avctx->bit_rate = s->bit_rate;

  2309.     avctx->sub_id = s->layer;

  2310. 

  2311.     if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))

  2312.         return -1;

  2313.     *data_size = 0;

  2314. 

  2315.     if(s->frame_size<=0 || s->frame_size > buf_size){

  2316.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  2317.         return -1;

  2318.     }else if(s->frame_size < buf_size){

  2319.         av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");

  2320.         buf_size= s->frame_size;

  2321.     }

  2322. 

  2323.     out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2324.     if(out_size>=0){

  2325.         *data_size = out_size;

  2326.         avctx->sample_rate = s->sample_rate;

  2327.         //FIXME maybe move the other codec info stuff from above here too

  2328.     }else

  2329.         av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed

  2330.     s->frame_size = 0;

  2331.     return buf_size;

  2332. }

  2333. 

  2334. static void flush(AVCodecContext *avctx){

  2335.     MPADecodeContext *s = avctx->priv_data;

  2336.     memset(s->synth_buf, 0, sizeof(s->synth_buf));

  2337.     s->last_buf_size= 0;

  2338. }

  2339. 

  2340. #if CONFIG_MP3ADU_DECODER

  2341. static int decode_frame_adu(AVCodecContext * avctx,

  2342.                         void *data, int *data_size,

  2343.                         AVPacket *avpkt)

  2344. {

  2345.     const uint8_t *buf = avpkt->data;

  2346.     int buf_size = avpkt->size;

  2347.     MPADecodeContext *s = avctx->priv_data;

  2348.     uint32_t header;

  2349.     int len, out_size;

  2350.     OUT_INT *out_samples = data;

  2351. 

  2352.     len = buf_size;

  2353. 

  2354.     // Discard too short frames

  2355.     if (buf_size < HEADER_SIZE) {

  2356.         *data_size = 0;

  2357.         return buf_size;

  2358.     }

  2359. 

  2360. 

  2361.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2362.         len = MPA_MAX_CODED_FRAME_SIZE;

  2363. 

  2364.     // Get header and restore sync word

  2365.     header = AV_RB32(buf) | 0xffe00000;

  2366. 

  2367.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2368.         *data_size = 0;

  2369.         return buf_size;

  2370.     }

  2371. 

  2372.     ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  2373.     /* update codec info */

  2374.     avctx->sample_rate = s->sample_rate;

  2375.     avctx->channels = s->nb_channels;

  2376.     avctx->bit_rate = s->bit_rate;

  2377.     avctx->sub_id = s->layer;

  2378. 

  2379.     s->frame_size = len;

  2380. 

  2381.     if (avctx->parse_only) {

  2382.         out_size = buf_size;

  2383.     } else {

  2384.         out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2385.     }

  2386. 

  2387.     *data_size = out_size;

  2388.     return buf_size;

  2389. }

  2390. #endif /* CONFIG_MP3ADU_DECODER */

  2391. 

  2392. #if CONFIG_MP3ON4_DECODER

  2393. 

  2394. /**

  2395.  * Context for MP3On4 decoder

  2396.  */

  2397. typedef struct MP3On4DecodeContext {

  2398.     int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)

  2399.     int syncword; ///< syncword patch

  2400.     const uint8_t *coff; ///< channels offsets in output buffer

  2401.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  2402. } MP3On4DecodeContext;

  2403. 

  2404. #include "mpeg4audio.h"

  2405. 

  2406. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  2407. static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */

  2408. /* offsets into output buffer, assume output order is FL FR BL BR C LFE */

  2409. static const uint8_t chan_offset[8][5] = {

  2410.     {0},

  2411.     {0},            // C

  2412.     {0},            // FLR

  2413.     {2,0},          // C FLR

  2414.     {2,0,3},        // C FLR BS

  2415.     {4,0,2},        // C FLR BLRS

  2416.     {4,0,2,5},      // C FLR BLRS LFE

  2417.     {4,0,2,6,5},    // C FLR BLRS BLR LFE

  2418. };

  2419. 

  2420. 

  2421. static int decode_init_mp3on4(AVCodecContext * avctx)

  2422. {

  2423.     MP3On4DecodeContext *s = avctx->priv_data;

  2424.     MPEG4AudioConfig cfg;

  2425.     int i;

  2426. 

  2427.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  2428.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  2429.         return -1;

  2430.     }

  2431. 

  2432.     ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);

  2433.     if (!cfg.chan_config || cfg.chan_config > 7) {

  2434.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  2435.         return -1;

  2436.     }

  2437.     s->frames = mp3Frames[cfg.chan_config];

  2438.     s->coff = chan_offset[cfg.chan_config];

  2439.     avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];

  2440. 

  2441.     if (cfg.sample_rate < 16000)

  2442.         s->syncword = 0xffe00000;

  2443.     else

  2444.         s->syncword = 0xfff00000;

  2445. 

  2446.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  2447.      * We replace avctx->priv_data with the context of the first decoder so that

  2448.      * decode_init() does not have to be changed.

  2449.      * Other decoders will be initialized here copying data from the first context

  2450.      */

  2451.     // Allocate zeroed memory for the first decoder context

  2452.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  2453.     // Put decoder context in place to make init_decode() happy

  2454.     avctx->priv_data = s->mp3decctx[0];

  2455.     decode_init(avctx);

  2456.     // Restore mp3on4 context pointer

  2457.     avctx->priv_data = s;

  2458.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  2459. 

  2460.     /* Create a separate codec/context for each frame (first is already ok).

  2461.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  2462.      */

  2463.     for (i = 1; i < s->frames; i++) {

  2464.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  2465.         s->mp3decctx[i]->adu_mode = 1;

  2466.         s->mp3decctx[i]->avctx = avctx;

  2467.     }

  2468. 

  2469.     return 0;

  2470. }

  2471. 

  2472. 

  2473. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  2474. {

  2475.     MP3On4DecodeContext *s = avctx->priv_data;

  2476.     int i;

  2477. 

  2478.     for (i = 0; i < s->frames; i++)

  2479.         if (s->mp3decctx[i])

  2480.             av_free(s->mp3decctx[i]);

  2481. 

  2482.     return 0;

  2483. }

  2484. 

  2485. 

  2486. static int decode_frame_mp3on4(AVCodecContext * avctx,

  2487.                         void *data, int *data_size,

  2488.                         AVPacket *avpkt)

  2489. {

  2490.     const uint8_t *buf = avpkt->data;

  2491.     int buf_size = avpkt->size;

  2492.     MP3On4DecodeContext *s = avctx->priv_data;

  2493.     MPADecodeContext *m;

  2494.     int fsize, len = buf_size, out_size = 0;

  2495.     uint32_t header;

  2496.     OUT_INT *out_samples = data;

  2497.     OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];

  2498.     OUT_INT *outptr, *bp;

  2499.     int fr, j, n;

  2500. 

  2501.     if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))

  2502.         return -1;

  2503. 

  2504.     *data_size = 0;

  2505.     // Discard too short frames

  2506.     if (buf_size < HEADER_SIZE)

  2507.         return -1;

  2508. 

  2509.     // If only one decoder interleave is not needed

  2510.     outptr = s->frames == 1 ? out_samples : decoded_buf;

  2511. 

  2512.     avctx->bit_rate = 0;

  2513. 

  2514.     for (fr = 0; fr < s->frames; fr++) {

  2515.         fsize = AV_RB16(buf) >> 4;

  2516.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  2517.         m = s->mp3decctx[fr];

  2518.         assert (m != NULL);

  2519. 

  2520.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  2521. 

  2522.         if (ff_mpa_check_header(header) < 0) // Bad header, discard block

  2523.             break;

  2524. 

  2525.         ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  2526.         out_size += mp_decode_frame(m, outptr, buf, fsize);

  2527.         buf += fsize;

  2528.         len -= fsize;

  2529. 

  2530.         if(s->frames > 1) {

  2531.             n = m->avctx->frame_size*m->nb_channels;

  2532.             /* interleave output data */

  2533.             bp = out_samples + s->coff[fr];

  2534.             if(m->nb_channels == 1) {

  2535.                 for(j = 0; j < n; j++) {

  2536.                     *bp = decoded_buf[j];

  2537.                     bp += avctx->channels;

  2538.                 }

  2539.             } else {

  2540.                 for(j = 0; j < n; j++) {

  2541.                     bp[0] = decoded_buf[j++];

  2542.                     bp[1] = decoded_buf[j];

  2543.                     bp += avctx->channels;

  2544.                 }

  2545.             }

  2546.         }

  2547.         avctx->bit_rate += m->bit_rate;

  2548.     }

  2549. 

  2550.     /* update codec info */

  2551.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  2552. 

  2553.     *data_size = out_size;

  2554.     return buf_size;

  2555. }

  2556. #endif /* CONFIG_MP3ON4_DECODER */

  2557. 

  2558. #if !CONFIG_FLOAT

  2559. #if CONFIG_MP1_DECODER

  2560. AVCodec mp1_decoder =

  2561. {

  2562.     "mp1",

  2563.     AVMEDIA_TYPE_AUDIO,

  2564.     CODEC_ID_MP1,

  2565.     sizeof(MPADecodeContext),

  2566.     decode_init,

  2567.     NULL,

  2568.     NULL,

  2569.     decode_frame,

  2570.     CODEC_CAP_PARSE_ONLY,

  2571.     .flush= flush,

  2572.     .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),

  2573. };

  2574. #endif

  2575. #if CONFIG_MP2_DECODER

  2576. AVCodec mp2_decoder =

  2577. {

  2578.     "mp2",

  2579.     AVMEDIA_TYPE_AUDIO,

  2580.     CODEC_ID_MP2,

  2581.     sizeof(MPADecodeContext),

  2582.     decode_init,

  2583.     NULL,

  2584.     NULL,

  2585.     decode_frame,

  2586.     CODEC_CAP_PARSE_ONLY,

  2587.     .flush= flush,

  2588.     .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),

  2589. };

  2590. #endif

  2591. #if CONFIG_MP3_DECODER

  2592. AVCodec mp3_decoder =

  2593. {

  2594.     "mp3",

  2595.     AVMEDIA_TYPE_AUDIO,

  2596.     CODEC_ID_MP3,

  2597.     sizeof(MPADecodeContext),

  2598.     decode_init,

  2599.     NULL,

  2600.     NULL,

  2601.     decode_frame,

  2602.     CODEC_CAP_PARSE_ONLY,

  2603.     .flush= flush,

  2604.     .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),

  2605. };

  2606. #endif

  2607. #if CONFIG_MP3ADU_DECODER

  2608. AVCodec mp3adu_decoder =

  2609. {

  2610.     "mp3adu",

  2611.     AVMEDIA_TYPE_AUDIO,

  2612.     CODEC_ID_MP3ADU,

  2613.     sizeof(MPADecodeContext),

  2614.     decode_init,

  2615.     NULL,

  2616.     NULL,

  2617.     decode_frame_adu,

  2618.     CODEC_CAP_PARSE_ONLY,

  2619.     .flush= flush,

  2620.     .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),

  2621. };

  2622. #endif

  2623. #if CONFIG_MP3ON4_DECODER

  2624. AVCodec mp3on4_decoder =

  2625. {

  2626.     "mp3on4",

  2627.     AVMEDIA_TYPE_AUDIO,

  2628.     CODEC_ID_MP3ON4,

  2629.     sizeof(MP3On4DecodeContext),

  2630.     decode_init_mp3on4,

  2631.     NULL,

  2632.     decode_close_mp3on4,

  2633.     decode_frame_mp3on4,

  2634.     .flush= flush,

  2635.     .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),

  2636. };

  2637. #endif

  2638. #endif